Ownbox The Ownbox Foundation · Ownbox.io · Download PDF

The Ownbox Foundation

Two hundred hours of engineering, running on day one, owned by your company.

This page describes custom Ownbox implementation services, scoped and priced separately. It is not the standard self-service Ownbox Core or Ownbox Pro purchase.

Most AI engagements begin by signing a company up for a stack of tools, each billed monthly, each holding a piece of its data, none of them its own. Ownbox begins differently. Our forward-deployed engineer arrives with a built machine: one system, on a server your company owns, already running the groundwork every AI build needs. Your budget goes on the problem you brought us in to solve.

Already built: ~200 hours of engineering, done before day one
Owned by: Your company its server, its data, its AI account
Running: Day one minutes after the order

Why we arrive with a machine

A consultant with nothing but advice leaves a company renting. We leave it owning an asset: the platform, the work built on it, and the data inside it.

An asset, not a stack of subscriptions. The groundwork is paid for once, already tested, and already running before the first hour of your project.

What our engineer does with it

Stands it up on your company's own server and AI account
Builds the machines your business needs on top of it
Keeps all of that work in a folder our updates never touch
Leaves a build guide, a stable toolkit and a checker, so your team keeps going

We bring the machine. You keep it.

What your company keeps

The machine, and every machine built on it
All of its data, exportable to CSV and JSON on any day
Its AI, on its own Claude or ChatGPT account
A copy it can move to its own cloud account
A system that keeps running, with or without a support plan

The four that take a good developer 50 hours on their own

Admin console 16 h

Personal sign-in, owner and member roles, invitations and settings in four plain groups, designed for a mobile device first.

Your own AI 12 h

Your company's Claude or ChatGPT plan, or an Anthropic API key. Checked before it is saved, metered on every request, under a hard monthly ceiling.

MCP server 14 h

A secure door for AI agents: standard OAuth or a key, two permission levels, every action recorded. No connected agent can send, publish or spend.

Mobile app 10 h

Installs from the browser on iPhone and Android, wearing your company's name and logo, with a printable install guide for staff.

And the other 150 hours

Security and ownership 28 h

Single-tenant: shares nothing with any other company
Automatic HTTPS, for approved addresses only
Signed releases, checked against a trust file on the box
Downgrades and unknown signers refused
Only a root key can change who may sign
Separate keys for the console, the job queue and updates
Cross-site protection on every change
Passwords hashed, sign-in attempts throttled
Credentials never shown, exported or logged
Our setup key deleted at first boot
Server access with your own SSH key
Every dependency pinned and verified
Background services locked down

Updates that can't break you 20 h

Signed updates checked twice a day
Spread out, never the whole fleet at once
Undone by itself if not healthy in two minutes
Database snapshot before every change
Never follows an unreviewed branch
Pauses if the box's own files were edited
One button saves those edits and restores the files
An Updates page: version and last check
Every update decision recorded on the box

Reliability 24 h

Watchdog every 30 minutes: AI, disk, web, worker
A job queue that never runs a job twice
Stuck work re-queued, never lost
Every job's outcome recorded
Nightly backup, seven days kept
Continuous off-site backup, ready to switch on
A restore drill and a self-check tool
Back by itself after a restart
A health check that names the exact version
A broken add-on never takes the box down
A signed six-hourly check-in: health, plan, head count, no customer data. Can be switched off

AI you control and can afford 16 h

One gateway for every AI request
Work that needs no AI never pays for it
Cost checked against the ceiling before each request
Every request logged against the job behind it
Reused context cached to cost less
A lighter model for simple tasks
A plan's usage limit pauses work, never fails it
Automatic retry when a provider is busy
A knowledge folder rides along on every request
Paid outside services held to their own caps

Your team 12 h

Invite, revoke and restore people on one page
Invitees set their own password
A revoked person is out on their next click
Members see only what they can use
One place for every setting, with who changed it
Light, dark or automatic, per person
A stop-everything switch for the owner
The system wears your company's name
Type and tap targets sized for a mobile device

The Morning Review 14 h

Built every 15 minutes, every day kept
A section for each machine; a quiet day stays quiet
Owner-only spend meters
By email, through your own Resend or SMTP account
Sent at most once a day, even if retried
By Slack when connected
Emails only your own people, never your customers
Your AI can read it over MCP

Built to grow 26 h

A folder of your own that updates never touch
Each new machine in one folder
A stable SDK, version 1, promised not to break
ownbox new writes a working starter machine
ownbox check flags anything unsafe, with the fix
A build guide for people and for AI coders
Machines add their own tools for your AI
A knowledge folder and settings file of your own
An update that stops a machine says which one, and whose fix it is
Export everything to CSV and JSON
Its own tests ship with it

Delivered running 10 h

Built automatically when ordered
Server, address and HTTPS set up
Running about five minutes after checkout
Handed to its owner with a one-time code
A set-up checklist that disappears when done

52 + 150 = about 200 hours of a good developer's time, before the first hour of your project. Conservative estimates, part by part.